Xiaomi's first open-source large reasoning model surpasses OpenAI's o1-mini in a parameter scale of 7 billion.

Today, Xiaomi has open-sourced its first large model "Xiaomi MiMo" designed for reasoning, enhancing reasoning capabilities through a combination of pre-training and post-training. In the field of mathematical reasoning and code competition evaluation sets, MiMo has surpassed OpenAI's proprietary reasoning model o1-mini and Alibaba's larger-scale open-source reasoning model QwQ-32B-Preview with only 7B parameters.